ChatGPT 5 feedback

Feedback
1

Hello, I’ve been using GPT-5 (testing all available GPT-5 models) since its release on Cursor, and I’ve noticed several important points:

  • Unlike GPT-4, GPT-5 lacks the ability to check file and image uploads.

  • It tends to generate responses quickly without much deep thinking, with about a 50% success rate.

  • Its planning capabilities are limited; it can only handle small tasks before it starts deviating and creating its own direction.

  • I rely more on GPT-4o because it can actually interpret file uploads and images, whereas GPT-5 only works with data that’s already been uploaded.

  • GPT-5 excels at resolving problems but sometimes introduces bigger issues by insisting on doing things its own way, even after a solution is found.

  • Occasionally (though not often), it claims to have made changes when in fact it has done nothing.

  • Sometimes GPT-5 attempts to “fix” something but doesn’t actually resolve the issue.

3 Likes
2

You rely on GPT 4o for coding ??
I suggest you try o3 who beats them all.

1 Like
3

Same GPT is DUMB.

4

That’s a bit hard, but I agree. GPT 5 messed up my code pretty bad.
In other case it was able to refactor huge file very well, where I think Claude 4 would fail.
So with fresh canvas and as senior advisor GPT 5 will be great, but letting it into existing codebase end up in disaster for me.
I have paid projects in production. I am far from senior dev, but I am far from pure vibe coder as well. Just some background on epx :).

5

GPT-5 ■■■■■ too much. It failed on a code review, despite Grok successfully doing the same code review. Attaching a video for reference where it suggested random things that don’t make sense to my codebase and went in a big “end” loop.

1 Like